The IBM RS/6000 SP Home Page
The IBM RS/6000 SP Resource Center
IBM Advanced Computing Technology Center
Announcement
Welcome
Welcome to the CSCAMM research dawson SP system! The CSCAMM
dawson cluster is a distributed-memory parallel system consisting
of one frame of total 8 wide IBM SP nodes, all nodes are running
the same version of AIX operating system (AIX 5.1).
User questions and problems should be directed to CSCAMM system staff by
contacting CSCAMM's System Administrator or by email .
CSCAMM SysAdmin
RM 4144 CSIC Building
301-405-8923
In September of 2001, Center for Scientific Computation and Mathematical
Modeling at University of Maryland acquired an
IBM RS/6000 Scalable POWERparallel System(SP) for research computing.
It came with 8 4-cpu Power3-II wide nodes.
one 1-cpu RS/6000 44p series model 170 for control workstation and
one 2-cpu 44p series 270 for front-end machine.
All nodes on the SP frame are interconnected with SP high-speed switch
MX2. Its peak data throughput is 1.5GB/second.
The SP system is also connected to the Internet through GigaBit ethernet
network.
All SP nodes are accessible through secure shell client. IBM compilers,
and several public domain software are available on the SP.
In addition, IBM software products designed to use the SP switch
for messages transfer are installed, including Parallel ESSL,
Distributed Debugger and Parallel Environment. These products offer
tools for compiling, running, monitoring, profiling, tuning and
debugging parallel programs.
The SP nodes process batch jobs via the IBM Loadleveler load-balancing
batch queueing system. Both serial and parallel jobs may
be submitted to the dawson SP nodes. Interactive use of the dawson
SP is allowed, but a limit of 20 minutes of elapsed CPU time per
interactive process is strictly enforced.
The following section will introduce the basic information for the dawson
cluster. Please send email to
if you have additional question or would like to have
something else to appear here.
Request An Account on Dawson SP
Please send email to
for more information.
The approval of such request requires a brief research plan which highlights
the scope and relevance of the usage of the SP2 machine and supporting
documentation by the chair/director of the unit submitting the request for
such an account.
User Home Directory and other user file system
We have 16x32 GB SSA disks attached
to dawson08 and dawson07 (also attached to dawson06 for failover if dawson08 or
dawson07 fails).
These disks are divided into 6 RAID-5 arrays and run in parallel using the
General Parellel File System (GPFS). All nodes will be able to access
the /gpfs just like accessing a local file system. The files stored in
GPFS striped across each RAID (which in turn are on two nodes). The data path
is through high performance IBM SP switch connecting all nodes. It is
not uncommon to achieve a sustained data transfer rate of 40 MB/sec or faster for files
in GPFS .
Job Scheduled System
The dawson SP system uses IBM job schedule
system LoadLeveler to manage user batch submission and execution.
Based on the
user inputs of various batch job requirements, the following job queue
classes have been defined for dawson SP system.
Class Name
Maximum Nodes
CPU Limit (hours)
Limit
pa
4
1
anytime
pb
4
4
midnight --- 8 am
pc
2
4
pd
2
8
pe
1
4
pf
1
8
sa (restricted)
1 processor
24
8 serial jobs (2 node total)
ia
20 minutes
interactive job
Submitting Batch Jobs to LoadLeveler
Before you can submit jobs to LoadLeveler, please add the following
path to your PATH: /usr/lpp/LoadL/full/bin, for man pages please
add the man path to MANPATH: /usr/lpp/LoadL/full/man . All loadleverler
commands are under /usr/lpp/LoadL/full/bin .
